Simplified Molecular Input Line Entry System Research Articles

Perplexity-Based Molecule Ranking and Bias Estimation of Chemical Language Models.

Chemical language models (CLMs) can be employed to design molecules with desired properties. CLMs generate new chemical structures in the form of textual representations, such as the simplified molecular input line entry system (SMILES) strings. However, the quality of these de novo generated molecules is difficult to assess a priori. In this study, we apply the perplexity metric to determine the degree to which the molecules generated by a CLM match the desired design objectives. This model-intrinsic score allows identifying and ranking the most promising molecular designs based on the probabilities learned by the CLM. Using perplexity to compare “greedy” (beam search) with “explorative” (multinomial sampling) methods for SMILES generation, certain advantages of multinomial sampling become apparent. Additionally, perplexity scoring is performed to identify undesired model biases introduced during model training and allows the development of a new ranking system to remove those undesired biases.

A review of molecular representation in the age of machine learning

AbstractResearch in chemistry increasingly requires interdisciplinary work prompted by, among other things, advances in computing, machine learning, and artificial intelligence. Everyone working with molecules, whether chemist or not, needs an understanding of the representation of molecules in a machine‐readable format, as this is central to computational chemistry. Four classes of representations are introduced: string, connection table, feature‐based, and computer‐learned representations. Three of the most significant representations are simplified molecular‐input line‐entry system (SMILES), International Chemical Identifier (InChI), and the MDL molfile, of which SMILES was the first to successfully be used in conjunction with a variational autoencoder (VAE) to yield a continuous representation of molecules. This is noteworthy because a continuous representation allows for efficient navigation of the immensely large chemical space of possible molecules. Since 2018, when the first model of this type was published, considerable effort has been put into developing novel and improved methodologies. Most, if not all, researchers in the community make their work easily accessible on GitHub, though discussion of computation time and domain of applicability is often overlooked. Herein, we present questions for consideration in future work which we believe will make chemical VAEs even more accessible.This article is categorized under: Data Science > Chemoinformatics

A surrogate model of sigma profile and COSMOSAC activity coefficient predictions of using transformer with SMILES input

COSMOSAC is a model that allows apriori predictions of activity coefficients for characterizing solute-solvent interactions. The method requires the input of sigma profile, the charge distribution on the surface of the molecules, which can be obtained through quantum mechanics calculation. Since Sigma profile is a unique function of molecular structure, it is desirable that they can be obtained using a surrogate model of the quantum computation with a molecular description as input. Previously, a model, the Universal Digital Chemical Space (UDCS), that was developed that allowed us to calculate the Sigma profiles used Simplified Molecular-Input-Line-Entry system (SMILES) as input. In this work, an improved version of this approach was developed using a Transformer model to encode the SMILES text string. Successive input elements in the text string, known as K-mers was also encoded and errors of predicted moments of Sigma profiles and prediction of activity coefficient of reference solvents were also considered as in the loss function. Results showed that while the prediction accuracy of Sigma profile (coefficient of determination R2) were not significantly improved, prediction accuracy of the first and second moment, especially the poorer ranked results; as well as the activity coefficients can be significantly improved with the inclusion of higher K-mers. Further improvement can be achieved with the inclusion of activity loss which substantially improved the accuracy of the 5th and 25th percentile of the moment loss and the activity coefficient of the species in n-hexane.

Role of the Backbone when Optimizing Functional Groups─A Theoretical Study Based on an Improved Inverse-Design Approach.

We present an improved inverse-design approach for automatically identifying molecular (or other) systems with optimal values for prechosen properties. The new approach uses SMILES (simplified molecular input line entry system) to describe molecular structures efficiently, a genetic algorithm to optimize the molecules automatically, and the DFTB+ (self-consistent charge density functional tight-binding) method to calculate electronic properties. Thereby, almost every class of materials─even macromolecules or monomers─can be studied easily. Without crossover operators but with only mutation operators, the genetic algorithm is more adaptive to SMILES while keeping its efficiency. DFTB+ is more accurate than the DFTB method used in our previous inverse-design approach for the study of excited states and charge transfer processes. The improved approach is applied to optimize benzene, pyridine, pyridazine, pyrimidine, and pyrazine derivatives for seven electronic properties, which all are highly relevant and important for the performance of molecules in solar cells. We found that for some electronic properties, the precise composition and structure of the backbone have remarkable impacts on the value of the electronic properties and/or on the set of functional groups that leads to the best performance. On the contrary, for other properties, these effects are less pronounced. The reasonable optimal functional groups and/or substitution patterns are reported.

VenomPred: A Machine Learning Based Platform for Molecular Toxicity Predictions.

The use of in silico toxicity prediction methods plays an important role in the selection of lead compounds and in ADMET studies since in vitro and in vivo methods are often limited by ethics, time, budget and other resources. In this context, we present our new web tool VenomPred, a user-friendly platform for evaluating the potential mutagenic, hepatotoxic, carcinogenic and estrogenic effects of small molecules. VenomPred platform employs several in-house Machine Learning (ML) models developed with datasets derived from VEGA QSAR, a software that includes a comprehensive collection of different toxicity models and has been used as a reference for building and evaluating our ML models. The results showed that our models achieved equal or better performance than those obtained with the reference models included in VEGA QSAR. In order to improve the predictive performance of our platform, we adopted a consensus approach combining the results of different ML models, which was able to predict chemical toxicity better than the single models. This improved method was thus implemented in the VenomPred platform, a freely accessible webserver that takes the SMILES (Simplified Molecular-Input Line-Entry System) strings of the compounds as input and sends the prediction results providing a probability score about their potential toxicity.

Deep Sequence Models for Ligand-Based Virtual Screening

The past few years have witnessed machine learning techniques take the limelight in multiple research domains. One such domain that has reaped the benefits of machine learning is computer-aided drug discovery, where the search space for candidate drug molecules is decreased using methods such as virtual screening. Current state-of-the-art sequential neural network models have shown promising results and we would like to replicate similar results with virtual screening using the encoded molecular information known as simplified molecular-input line-entry system (SMILES). Our work includes the use of attention-based sequential models — the long short-term memory with attention and an optimized version of the transformer network specifically designed to deal with SMILES (ChemBERTa). We also propose the “Overall Screening Efficacy”, an averaging metric that aggregates and encapsulates the model performance over multiple datasets. We found an overall improvement of about [Formula: see text] over the benchmark model, which relied on parallelized random forests.

An AI-based Prediction Model for Drug-drug Interactions in Osteoporosis and Paget's Diseases from SMILES.

The skeleton is one of the most important organs in the human body in assisting our motion and activities; however, bone density attenuates gradually as we age. Among common bone diseases are osteoporosis and Paget's, two of the most frequently found diseases in the elderly. Nowadays, a combination of multiple drugs is the optimal therapy to decelerate osteoporosis and Paget's pathologic process, which comes with various underlying adverse effects due to drug-drug interactions (DDIs). Artificial intelligence (AI) has the potential to evaluate the interaction, pharmacodynamics, and possible side effects between drugs. In this research, we created an AI-based machine-learning model to predict the outcomes of interactions between drugs used for osteoporosis and Paget's treatment, which helps mitigate the cost and time to implement the best combination of medications in clinical practice. In this study, a DDI dataset was collected from the DrugBank database within the osteoporosis and Paget diseases. We then extracted a variety of chemical features from the simplified molecular-input line-entry system (SMILES) of defined drug pairs that interact with each other. Finally, machine-learning algorithms were implemented to learn the extracted features. Our stack ensemble model from Random Forest and XGBoost reached an average accuracy of 74 % in predicting DDIs. It was superior to individual models as well as previous methods in terms of most measurement metrics. This study showed the potential of AI models in predicting DDIs of Osteoporosis-Paget's disease in particular, and other diseases in general.

The development of nano-QSPR models for viscosity of nanofluids using the index of ideality of correlation and the correlation intensity index

Utilizing nanofluids as a suspension containing nanoparticles in an ordinary liquid is a relatively new field, which received great attention in recent years. The main object of this investigation is focused on modeling the effective viscosity of nanofluids using the nano-quantitative structure-property relationship (nano-QSPR) paradigm. Two distinct data sets were considered containing four types of nanoparticles (Al2O3, CuO, SiO2, and ZnO) dispersed in water (as the most common base fluid) at volume fraction ranges of 1–5% and various shapes (blades, bricks, cylindrical, spherical, and platelets). Simplified Molecular Input-Line Entry System (SMILES) is a tool to represent the molecular structure. Quasi-SMILES is a sequence of symbols that represents all available data e.g. molecular structure together with physicochemical conditions. Taking into account the capability of quasi-SMILES molecular representation to define the eclectic data such as size and shape of nanoparticles, this notation was chosen to exemplify nanofluids structure. It is remarkable to point out that the proposed attitude to generate nano-QSPR models introduced a comparison between two specific predictive potential criteria using by Monte Carlo technique. It was concluded that the development of models based on Correlation intensity index (CII) is statistically more reliable than model generation based on the Index of ideality of correlation (IIC).

Deep Learning-Assisted Repurposing of Plant Compounds for Treating Vascular Calcification: An In Silico Study with Experimental Validation.

Background Vascular calcification (VC) constitutes subclinical vascular burden and increases cardiovascular mortality. Effective therapeutics for VC remains to be procured. We aimed to use a deep learning-based strategy to screen and uncover plant compounds that potentially can be repurposed for managing VC. Methods We integrated drugome, interactome, and diseasome information from Comparative Toxicogenomic Database (CTD), DrugBank, PubChem, Gene Ontology (GO), and BioGrid to analyze drug-disease associations. A deep representation learning was done using a high-level description of the local network architecture and features of the entities, followed by learning the global embeddings of nodes derived from a heterogeneous network using the graph neural network architecture and a random forest classifier established for prediction. Predicted results were tested in an in vitro VC model for validity based on the probability scores. Results We collected 6,790 compounds with available Simplified Molecular-Input Line-Entry System (SMILES) data, 11,958 GO terms, 7,238 diseases, and 25,482 proteins, followed by local embedding vectors using an end-to-end transformer network and a node2vec algorithm and global embedding vectors learned from heterogeneous network via the graph neural network. Our algorithm conferred a good distinction between potential compounds, presenting as higher prediction scores for the compound categories with a higher potential but lower scores for other categories. Probability score-dependent selection revealed that antioxidants such as sulforaphane and daidzein were potentially effective compounds against VC, while catechin had low probability. All three compounds were validated in vitro. Conclusions Our findings exemplify the utility of deep learning in identifying promising VC-treating plant compounds. Our model can be a quick and comprehensive computational screening tool to assist in the early drug discovery process.

Study of quantitative structure-property relationship for density of ionic liquids based on Monte Carlo optimization

Ionic liquids (ILs) have attracted increasing interests and applications due to its unique physiochemical properties. Density is a vital physical property of ILs. In this work, a comprehensive collection of density data is conducted on 184 variable ILs. The study of quantitative structure-property relationship (QSPR) is carried out for the selected density data of ILs using simplified molecular input line entry specification (SMILES) as the representation of the molecular structure of ILs by means of CORAL software. QSPR relationships were constructed with the balance of correlations (BC) and the classic scheme. Results from three random splits displayed desirable models for predicting the external test set with the correlation coefficient (R2) and cross validated correlation coefficient (Q2) in ranges of 0.8234–0.9770 and 0.7599–0.9745, respectively. The best predictions obtained by the balance of correlations along with the global SMILES descriptors are included in the modeling process. The average statistical characteristics of the external test set are the following: n =36, R2 =0.9770, Q2= 0.9745, standard error of estimation (s)=0.023, mean absolute error (MAE) =0.018 and Fischer F-ratio (F)=1443

Pushing the Boundaries of Molecular Property Prediction for Drug Discovery with Multitask Learning BERT Enhanced by SMILES Enumeration

Accurate prediction of pharmacological properties of small molecules is becoming increasingly important in drug discovery. Traditional feature-engineering approaches heavily rely on handcrafted descriptors and/or fingerprints, which need extensive human expert knowledge. With the rapid progress of artificial intelligence technology, data-driven deep learning methods have shown unparalleled advantages over feature-engineering-based methods. However, existing deep learning methods usually suffer from the scarcity of labeled data and the inability to share information between different tasks when applied to predicting molecular properties, thus resulting in poor generalization capability. Here, we proposed a novel multitask learning BERT (Bidirectional Encoder Representations from Transformer) framework, named MTL-BERT, which leverages large-scale pre-training, multitask learning, and SMILES (simplified molecular input line entry specification) enumeration to alleviate the data scarcity problem. MTL-BERT first exploits a large amount of unlabeled data through self-supervised pretraining to mine the rich contextual information in SMILES strings and then fine-tunes the pretrained model for multiple downstream tasks simultaneously by leveraging their shared information. Meanwhile, SMILES enumeration is used as a data enhancement strategy during the pretraining, fine-tuning, and test phases to substantially increase data diversity and help to learn the key relevant patterns from complex SMILES strings. The experimental results showed that the pretrained MTL-BERT model with few additional fine-tuning can achieve much better performance than the state-of-the-art methods on most of the 60 practical molecular datasets. Additionally, the MTL-BERT model leverages attention mechanisms to focus on SMILES character features essential to target properties for model interpretability.

Drug repositioning of COVID-19 based on mixed graph network and ion channel.

Research on the relationship between drugs and targets is the key to precision medicine. Ion channel is a kind of important drug targets. Aiming at the urgent needs of corona virus disease 2019 (COVID-19) treatment and drug development, this paper designed a mixed graph network model to predict the affinity between ion channel targets of COVID-19 and drugs. According to the simplified molecular input line entry specification (SMILES) code of drugs, firstly, the atomic features were extracted to construct the point sets, and edge sets were constructed according to atomic bonds. Then the undirected graph with atomic features was generated by RDKit tool and the graph attention layer was used to extract the drug feature information. Five ion channel target proteins were screened from the whole SARS-CoV-2 genome sequences of NCBI database, and the protein features were extracted by convolution neural network (CNN). Using attention mechanism and graph convolutional network (GCN), the extracted drug features and target features information were connected. After two full connection layers operation, the drug-target affinity was output, and model was obtained. Kiba dataset was used to train the model and determine the model parameters. Compared with DeepDTA, WideDTA, graph attention network (GAT), GCN and graph isomorphism network (GIN) models, it was proved that the mean square error (MSE) of the proposed model was decreased by 0.055, 0.04, 0.001, 0.046, 0.013 and the consistency index (CI) was increased by 0.028, 0.016, 0.003, 0.03 and 0.01, respectively. It can predict the drug-target affinity more accurately. According to the prediction results of drug-target affinity of SARS-CoV-2 ion channel targets, seven kinds of small molecule drugs acting on five ion channel targets were obtained, namely SCH-47112, Dehydroaltenusin, alternariol 5-o-sulfate, LPA1 antagonist 1, alternariol, butin, and AT-9283.These drugs provide a reference for drug repositioning and precise treatment of COVID-19.

AMDE: a novel attention-mechanism-based multidimensional feature encoder for drug–drug interaction prediction

The properties of the drug may be altered by the combination, which may cause unexpected drug-drug interactions (DDIs). Prediction of DDIs provides combination strategies of drugs for systematic and effective treatment. In most of deep learning-based methods for predicting DDI, encoded information about the drugs is insufficient in some extent, which limits the performances of DDIs prediction. In this work, we propose a novel attention-mechanism-based multidimensional feature encoder for DDIs prediction, namely attention-based multidimensional feature encoder (AMDE). Specifically, in AMDE, we encode drug features from multiple dimensions, including information from both Simplified Molecular-Input Line-Entry System sequence and atomic graph of the drug. Data experiments are conducted on DDI data set selected from Drugbank, involving a total of 34282 DDI relationships with 17141 positive DDI samples and 17141 negative samples. Experimental results show that our AMDE performs better than some state-of-the-art baseline methods, including Random Forest, One-Dimension Convolutional Neural Networks, DeepDrug, Long Short-Term Memory, Seq2seq, Deepconv, DeepDDI, Graph Attention Networks and Knowledge Graph Neural Networks. In practice, we select a set of 150 drugs with 3723 DDIs, which are never appeared in training, validation and test sets. AMDE performs well in DDIs prediction task, with AUROC and AUPRC 0.981 and 0.975. As well, we use Torasemide (DB00214) as an example and predict the most likely drug to interact with it. The top 15 scores all have been reported with clear interactions in literatures.

Integration of System Biology Tools to Investigate Huperzine A as an Anti-Alzheimer Agent.

Aim: The present study aimed to investigate huperzine A as an anti-Alzheimer agent based on the principle that a single compound can regulate multiple proteins and associated pathways, using system biology tools. Methodology: The simplified molecular-input line-entry system of huperzine A was retrieved from the PubChem database, and its targets were predicted using SwissTargetPrediction. These targets were matched with the proteins deposited in DisGeNET for Alzheimer disease and enriched in STRING to identify the probably regulated pathways, cellular components, biological processes, and molecular function. Furthermore, huperzine A was docked against acetylcholinesterase using AutoDock Vina, and simulations were performed with the Gromacs package to take into account the dynamics of the system and its effect on the stability and function of the ligands. Results: A total of 100 targets were predicted to be targeted by huperzine A, of which 42 were regulated at a minimum probability of 0.05. Similarly, 101 Kyoto Encyclopedia of Genes and Genomes pathways were triggered, in which neuroactive ligand–receptor interactions scored the least false discovery rate. Also, huperzine A was predicted to modulate 54 cellular components, 120 molecular functions, and 873 biological processes. Furthermore, huperzine A possessed a binding affinity of −8.7 kcal/mol with AChE and interacted within the active site of AChE via H-bonds and hydrophobic interactions.

MERMAID: an open source automated hit-to-lead method based on deep reinforcement learning

The hit-to-lead process makes the physicochemical properties of the hit molecules that show the desired type of activity obtained in the screening assay more drug-like. Deep learning-based molecular generative models are expected to contribute to the hit-to-lead process. The simplified molecular input line entry system (SMILES), which is a string of alphanumeric characters representing the chemical structure of a molecule, is one of the most commonly used representations of molecules, and molecular generative models based on SMILES have achieved significant success. However, in contrast to molecular graphs, during the process of generation, SMILES are not considered as valid SMILES. Further, it is quite difficult to generate molecules starting from a certain molecule, thus making it difficult to apply SMILES to the hit-to-lead process. In this study, we have developed a SMILES-based generative model that can be generated starting from a certain molecule. This method generates partial SMILES and inserts it into the original SMILES using Monte Carlo Tree Search and a Recurrent Neural Network. We validated our method using a molecule dataset obtained from the ZINC database and successfully generated molecules that were both well optimized for the objectives of the quantitative estimate of drug-likeness (QED) and penalized octanol-water partition coefficient (PLogP) optimization. The source code is available at https://github.com/sekijima-lab/mermaid.

PSO-ELM with Modified Acceleration Coefficients for Classifying the Active Compound

Background: The classification of active compounds based on their function using machine learning is essential for predicting the function of new active compounds quickly. These classification results are beneficial to accelerate the work of laboratory assistants in identifying the function of active compounds. In this study, an active compound is represented by the Simplified Molecular-Input Line-Entry System (SMILES) code. Objective: This paper proposes a modified acceleration coefficient to improve the PSO-ELM performance for predicting the function of the SMILES code. Method: The research uses a machine-learning algorithm that is a combination of the Particle Swarm Optimization and Extreme Learning Machine (PSO-ELM). ELM is used to classify the SMILES code, while PSO is used to optimize ELM parameters, i.e., weight, bias, and the number of hidden neurons. The important parameters that significantly influence the PSO performance are acceleration coefficients. The acceleration coefficients, that are modified Sigmoid-Based Acceleration Coefficient (SBAC), are introduced and compared with seven other acceleration coefficients. Results: The experimental results show that the sensitivity, specificity, accuracy, and Area Under the Curve (AUC) of the proposed acceleration coefficients outperform all other acceleration coefficients. The increased accuracy of the proposed can reach up to 2.64%, 5.84%, 7.93%, 8.44%, and 16.29% for Support Vector Machine (SVM), decision tree, AdaBoost, MLP Classifier, and Gaussian Naïve Bayes algorithms, respectively. Conclusion: The acceleration coefficients affect the prediction accuracy of the SMILES code classification. The proposed acceleration coefficients improve the performance of the PSO-ELM for predicting the function of the SMILES code.

Machine Learning-Based Prediction of Drug-Drug Interactions for Histamine Antagonist Using Hybrid Chemical Features.

The requesting of detailed information on new drugs including drug-drug interactions or targets is often unavailable and resource-intensive in assessing adverse drug events. To shorten the common evaluation process of drug-drug interactions, we present a machine learning framework-HAINI to predict DDI types for histamine antagonist drugs using simplified molecular-input line-entry systems (SMILES) combined with interaction features based on CYP450 group as inputs. The data used in our research consisted of approved drugs of histamine antagonists that are connected to 26,344 DDI pairs from the DrugBank database. Various classification algorithms such as Naive Bayes, Decision Tree, Random Forest, Logistic Regression, and XGBoost were used with 5-fold cross-validation to approach a large-scale DDIs prediction among histamine antagonist drugs. The prediction performance shows that our model outperformed previously published works on DDI prediction with the best precision of 0.788, a recall of 0.921, and an F1-score of 0.838 among 19 given DDIs types. An important finding of the study is that our prediction is based solely on the SMILES and CYP450 and thus can be applied at the early stage of drug development.

CBDPS 1.0: A Python GUI Application for Machine Learning Models to Predict Bitter-Tasting Children’s Oral Medicines

Bitter tastes are innately aversive and are thought to help protect animals from consuming poisons. Children are extremely sensitive to drug tastes, and their compliance is especially poor with bitter medicine. Therefore, judging whether a drug is bitter and adopting flavor correction and taste-masking strategies are key to solving the problem of drug compliance in children. Although various machine learning models for bitterness and sweetness prediction have been reported in the literature, no learning model or bitterness database for children's medication has yet been reported. In this study, we trained four different machine learning models to predict bitterness. The goal of this study was to develop and validate a machine learning model called the "Children's Bitter Drug Prediction System" (CBDPS) based on Tkinter, which predicts the bitterness of a medicine based on its chemical structure. Users can enter the Simplified Molecular-Input Line-Entry System (SMILES) formula for a single compound or multiple compounds, and CBDPS will predict the bitterness of children's medicines made from those XGBoost-Molecular ACCess System (XgBoost-MACCS) model yielded an accuracy of 88% under cross-validation.

Prediction of power conversion efficiency of phenothiazine-based dye-sensitized solar cells using Monte Carlo method with index of ideality of correlation

ABSTRACT Simplified molecular-input line-entry system (SMILES) notation and inbuilt Monte Carlo algorithm of CORAL software were employed to construct generative and prediction QSPR models for the analysis of the power conversion efficiency (PCE) of 215 phenothiazine derivatives. The dataset was divided into four splits and each split was further divided into four sets. A hybrid descriptor, a combination of SMILES and hydrogen suppressed graph (HSG), was employed to build reliable and robust QSPR models. The role of the index of ideality of correlation (IIC) was also studied in depth. We performed a comparative study to predict PCE using two target functions (TF1 without IIC and TF2 with IIC). Eight QSPR models were developed and the models developed with TF2 was shown robust and reliable. The QSPR model generated from split 4 was considered a leading model. The different statistical benchmarks were computed for the lead model and these were ; ; ; ; ; ; ; ; IICtraining set = 0.8590; IICinvisible training set = 0.8297; IICcalibration set = 0.8796; IICvalidation set = 0.8293, etc. The promoters of increase and decrease of endpoint PCE were also extracted.

Application of quasi-SMILES to the model of gold-nanoparticles uptake in A549cells.

Cell death is critical to human health and is associated with a variety of medical conditions. Therefore, new controllers of cell death are needed for the treatment of diverse diseases. In particular, nanoparticles (NP) are now regularly used in various applications, including a variety of products and medicines. Gold nanoparticles (GNPs) are widely used in the medical field against A549 lung carcinoma cells. The present study is devoted to developing computational models of the cellular uptake potentials by A549cells of gold nanoparticles (GNPs) under various conditions. Simplified molecular input-line entry system (SMILES) is an efficient tool to represent the molecular structure by a sequence of symbols. Quasi-SMILES represents an extended version of SMILES where symbols to denote physicochemical and/or biochemical conditions are added. In other words, the quasi-SMILES represents a biochemical (medical) phenomenon related to the whole matter (not only molecular structure). We developed models for the cellular utpake potential of gold nanoparticles (GNPs) in A549 [10-11g Au/Cell] under various conditions based on quasi-SMILES using the Monte Carlo method. The statistical quality of these models is quite good.